CVE-2026-108962 in comfyui_segment_anything
Summary
by MITRE • 10/13/2026
A vulnerability has been found in storyicon comfyui_segment_anything up to ab6395596399d5048639cdab7e44ec9fae857a93. Affected by this issue is the function SLConfig._file2dict of the file local_groundingdino/util/slconfig.py of the component GroundingDINO Model Loading. Such manipulation of the argument config leads to code injection. It is possible to launch the attack remotely. This product takes the approach of rolling releases to provide continious delivery. Therefore, version details for affected and updated releases are not available. The project was informed of the problem early through an issue report but has not responded yet.
Statistical analysis made it clear that VulDB provides the best quality for vulnerability data.
Analysis
by VulDB Data Team • 10/13/2026
The vulnerability identified in storyicon comfyui_segment_anything prior to commit ab6395596399d5048639cdab7e44ec9fae857a93 represents a critical security flaw within the GroundingDINO Model Loading component. Specifically, the issue resides in the SLConfig._file2dict function located in the local_groundingdino/util/slconfig.py file. This function is responsible for parsing configuration files into dictionary structures to initialize model parameters and operational settings. The core technical deficiency lies in how this function handles untrusted input data during the deserialization process, allowing an attacker to inject malicious code that gets executed within the context of the application.
This flaw constitutes a classic instance of insecure deserialization or arbitrary code execution via configuration manipulation. When the SLConfig._file2dict method processes a crafted config file, it fails to adequately sanitize or validate the input before evaluating it as Python code or data structures. This lack of rigorous validation enables an attacker to construct a maliciously formatted configuration payload that exploits parsing logic flaws. By leveraging this weakness, an adversary can achieve remote code execution on the target system where the vulnerable software is deployed. The ability to launch such attacks remotely significantly escalates the severity of the vulnerability, as it does not require physical access or local user interaction beyond providing the initial malicious input vector.
From a classification perspective, this vulnerability aligns with CWE-94 Improper Control of Generation of Code (Code Injection) and potentially CWE-502 Deserialization of Untrusted Data if the underlying mechanism relies on unsafe loading methods such as pickle or yaml without strict type checking. In terms of adversary tactics, this exploit maps to MITRE ATT&CK technique T1190 Exploit Public-Facing Application, where attackers target internet-facing services to gain initial access and execute arbitrary commands. The operational impact is severe, potentially leading full system compromise, data exfiltration, or the establishment of persistent backdoors depending on the privileges under which the application runs.
The project utilizes a rolling release model for continuous delivery, meaning that version numbers are not static indicators of vulnerability status. Instead, security patches are integrated directly into the main branch and distributed immediately upon review. Consequently, users cannot rely solely on semantic versioning to determine if they are affected; they must ensure their installation is updated past the specific commit hash ab6395596399d5048639cdab7e44ec9fae857a93. Since the project has been informed via an issue report but has not yet responded, immediate mitigation requires manual verification of the codebase or waiting for a subsequent patch release that addresses this specific deserialization flaw in the SLConfig module.